Formal Analysis of Electronic Exams
Jannik Dreier
1
, Rosario Giustolisi
2
, Ali Kassem
3
,
Pascal Lafourcade
4
, Gabriele Lenzini
2
and Peter Y. A. Ryan
2
1
Institute of Information Security, ETH Zurich, Zurich, Switzerland
2
SnT/University of Luxembourg, Luxembourg, Luxembourg
3
Universit´e Grenoble Alpes, CNRS, VERIMAG, Grenoble, France
4
University d’Auvergne, LIMOS, Aubi`ere, France
Keywords:
Electronic Exams, Formal Verification, Authentication, Privacy, Applied π-Calculus, ProVerif.
Abstract:
Universities and other educational organizations are adopting computer and Internet-based assessment tools
(herein called e-exams) to reach widespread audiences. While this makes examination tests more accessible, it
exposes them to new threats. At present, there are very few strategies to check such systems for security, also
there is a lack of formal security definitions in this domain. This paper fills this gap: in the formal framework
of the applied π-calculus, we define several fundamental authentication and privacy properties and establish
the first theoretical framework for the security analysis of e-exam protocols. As proof of concept we analyze
two of such protocols with ProVerif. The first “secure electronic exam system” proposed in the literature turns
out to have several severe problems. The second protocol, called Remark!, is proved to satisfy all the security
properties assuming access control on the bulletin board. We propose a simple protocol modification that
removes the need of such assumption though guaranteeing all the security properties.
1 INTRODUCTION
Electronic exams (in short, e-exams) are computer-
based systems employed to assess the skills, the capa-
bilities or the knowledge of students and profession-
als. Their importance has raised considerably since
several educational and testing institutions began to
offer e-exams as a service open to a worldwide-spread
audience. For instance, universities like MIT, Stan-
ford and Berkeley have set them up in Massive Open
Online Course (MOOC). Students can be marked on-
line, although not yet granted with a legally valid
diploma. Other institutions employ e-exams to grant
students with certificates which have an officially rec-
ognized validity. This happens, for instance, with
testing organizations like ETS
2
, the world leader in
the business of assessing content-knowledge and abil-
ities in various subjects. Other examples are CISCO
and Microsoft’s career certification programs, and
ECDL, the pioneer in assessing people’s computer of-
fice skills and in releasing “European Computer Driv-
ing Licence”.
For all these and similar institutes, e-exam sys-
This research was conducted with the support of the
Digital trust Chair from the University of Auvergne Foun-
dation.
2
https://www.ets.org/
tems are a promise for a better and a cheaper orga-
nization and management of tests: e-exams are flexi-
ble in where and when exams can be set (Hjeltnes and
Hansson, 2005), their test sessions can be easily open
to a very large public of candidates and, if the imple-
mentation allows automatic marking, their results are
immediately available.
So far, the main concern about e-exam secu-
rity has been about student cheating and imperson-
ation (Weippl, 2005). Since such threats mainly come
from students, most institutions have arranged invig-
ilated testing. The invigilator can be a person, like a
lecturer attending at the test location, or a software,
like ProctorU
3
running on the candidate’s computer.
A proctor is meant to supervise the test and to detect
and report, perhaps also discourage, any attempt of
fraud. However, e-exams are threatened by more se-
rious problems than student cheating. As evidenced
by recent scandals, dishonest acts can come also from
other parties than candidates, such as bribed exam-
iners or misbehaving exam authorities. The conse-
quences are usually worse in these cases than in those
due to student cheating. In the Atlanta scandal, school
authorities colluded in changing student marks to im-
prove their institutions rankings and get more pub-
3
http://www.proctoru.com/
101
Dreier J., Giustolisi R., Kassem A., Lafourcade P., Lenzini G. and Y. A. Ryan P..
Formal Analysis of Electronic Exams.
DOI: 10.5220/0005050901010112
In Proceedings of the 11th International Conference on Security and Cryptography (SECRYPT-2014), pages 101-112
ISBN: 978-989-758-045-1
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
lic funds (Copeland, 2013). In a BBC investigation,
ETS was shown vulnerable to a fraud perpetrated by
official invigilators: in collusion with all the candi-
dates who were there to get their visas, the invigilators
dictated the correct answers during the test (Watson,
2014).
We keep a neutral position in the debate about
whether e-exams are actually a beneficial and promis-
ing choice in promotingand supporting education, but
it is a fact that the adoption of computer-based assess-
ment is increasing. It is also a fact that such a growth
has not been followed nor was it preceded by a rig-
orous understanding and analysis of security. There
is then a need for a formal framework to define and
analyse the security of e-exam protocols.
Contributions: We present the first formalization
for e-exams, and define several fundamental security
properties for e-exams. We categorize them in two
main classes: (a) authentication properties, includ-
ing Answer Origin Authentication, Form Authorship,
Form Authenticity, and Mark Authenticity, and (b) pri-
vacy properties, containing Question Indistinguisha-
bility, Anonymous Marking, Anonymous Examiner,
Mark Privacy, and Mark Anonymity. We develop our
formal framework in the applied π-calculus (Abadi
and Fournet, 2001), wherein we propose a model for
the typical e-exam’s processes and phases and define
all our properties.
We validate our approach by analyzing two e-
exam protocols. The first is an internet exam pro-
tocol proposed by Huszti et al. (Huszti and Peth˝o,
2010). The second is a recent protocol proposed by
Giustolisi et al. (Giustolisi et al., 2014). We model
both protocols in the applied π-calculus, and check
them against our properties using ProVerif (Blanchet,
2001; Blanchet et al., 2008). Our security analysis
reveals several weaknesses in the first protocol, even
without considering dishonest parties. We show that
the second protocol is secure if all parties are honest,
and we also consider the situation where some par-
ties are dishonest, i.e., collaborate with the attacker.
In this case we discover a weakness on Form Authen-
ticity, and we propose a simple fix to overcome this
weakness.
Beyond these results, this work is the first theo-
retical framework for the security analysis of e-exams
protocols. It can be generalized in a straightforward
way to study traditional exam protocols.
Related Work: Only a few papers propose e-exam
protocols that guarantee some security, mainly under
the assumption that some authority is trusted (Huszti
and Peth˝o, 2010; Castell`a-Roca et al., 2006; Herrera-
Joancomart´ı et al., 2004; Bella et al., 2011). Few
other works (Giustolisi et al., 2013; Furnell et al.,
1998; Weippl, 2005) list some relevant properties for
e-exams, yet only informally.
Arapinis et al. (Arapinis et al., 2013) propose
a cloud-based protocol for conference management
system that supports applications, evaluations, and
decisions. They identify and analyze a few privacy
properties (secrecy and unlinkability) that should
hold despite a malicious-but-cautious cloud, and they
prove, using ProVerif, that their protocol satisfy them.
To the best of our knowledge, no formal defi-
nitions have been given for the security properties
of e-exam systems. There are instead papers pre-
senting the formalization and verification of proper-
ties in domains that seem related to e-exams, namely
e-voting (Dreier et al., 2011; Dreier et al., 2012b;
Dreier et al., 2012a; Backes et al., 2008a; Delaune
et al., 2009; Delaune et al., 2006a) and e-auction sys-
tems (Dong et al., 2010; Dreier et al., 2013b; Dreier
et al., 2013a).
Some of the security properties therein studied re-
mind those we are presenting for e-exams. For in-
stance, Answer Origin Authentication is analogous to
voter and bidder authentication. Mark Privacy re-
minds ballot privacy and losing bids privacy. Yet,
there are fundamental differences. In e-exams, An-
swer Authorship should be preserved even in the pres-
ence of colluding candidates. Conversely, vote (bid)
authorship is not a problem for e-voting (e-auction),
in fact unlinkability between a voter (bidder) and her
vote (bid) is a desired property. An other important
property for e-exams is to keep exam questions secret
until the exam ends. We do not find such a property in
e-voting where the candidates are previously known
to the voters, and in e-auction where the goods to bid
for are previously known to the bidders. Moreover,
properties such as Anonymous Marking, meaning that
the examiners do not know whose copy they are grad-
ing, evaluates to a sort of fixed-term anonymity. This
property is meant to hold during the marking, but is
trivially falsified when the marks are assigned to the
candidates.
Outline: In Section 2, we model e-exam protocols
in the applied π-calculus. Then, we specify secu-
rity properties in Section 3. We validate our frame-
work by analysing the security of two e-exam proto-
cols (Huszti and Peth˝o, 2010) and (Giustolisi et al.,
2014) in Section 4 and 5. Finally, in Section 6, we
discuss our results and outline the future work.
SECRYPT2014-InternationalConferenceonSecurityandCryptography
102
2 MODELLING
We model e-exam protocols in the applied π-calculus,
a process calculus designed for the verification of
cryptographic protocols. To perform the automatic
protocol verification, we use ProVerif. This tool
uses a process description based on the applied π-
calculus, but has syntactical extensions and is en-
riched by events to check reachability and correspon-
dence properties. Besides, it can check equivalence
properties. We use events to define various authen-
tication properties, and we model privacy properties
as equivalence properties. Precisely, honest parties
are modeled as processes in the applied π-calculus.
These processes can exchange messages on public
or private channels, create keys or fresh random val-
ues and perform tests and cryptographic operations,
which are modeled as functions on terms with respect
to an equational theory describing their properties.
The attacker has complete control of the network,
except the private channels: he can eavesdrop, re-
move, substitute, duplicate and delay messages that
the parties are sending one another, and insert mes-
sages of his choice on the public channels (like the
Dolev-Yao attacker (Dolev and Yao, 1983)). To cap-
ture threats due to collusions and coercions, we as-
sume dishonest parties. They cooperate with the at-
tacker, revealing their secret data (e.g., secret keys) to
him, or taking orders from him (e.g., how to answer a
question). We model such dishonest parties as in Def-
inition 8 from (Delaune et al., 2006b): if the process P
is an honest party, then the process P
c
1
,c
2
is its dishon-
est version. This is a variant of P which shares with
the attacker channels c
1
and c
2
. Through c
1
, P
c
1
,c
2
sends all its inputs and freshly generated names (but
not other channel names). From c
2
, P
c
1
,c
2
receives
messages that can influence its behaviour. For more
details about the applied π-calculus, its standard re-
sults and all the definitions used in this paper, we re-
mind to the papers (Abadi and Fournet, 2001; De-
laune et al., 2006b). An e-exam system involves dif-
ferent parties, among which are the candidates who
sit for the exam; the examiners who mark the answers
submitted by the candidates; the question committee,
which prepare the exam questions; the exam authori-
ties, which conducts the exam, that is registrars, invig-
ilators, exam collectors, and a notification committee.
In some protocols, an authority can be responsible of
two or more roles.
Definition 1. (E-exam protocol). An e-exam proto-
col is a tuple (C, E, Q, A
1
, . . . , A
l
, ˜n
p
), where C is the
process executed by the candidates, E is the process
executed by the examiners, Q is the process executed
by the question commitee, A
i
s are the processes ex-
ecuted by the authorities, and ˜n
p
is the set of private
channel names.
Note that all candidates and all examiners execute
the same process, but with different variable values,
e.g., keys, identities, and answers.
Definition 2. (E-exam instance). Given an e-
exam protocol an e-exam instance is a closed pro-
cess EP = ν˜n.(Cσ
id
1
σ
a
1
|. . . |Cσ
id
j
σ
a
j
|Eσ
id
1
σ
m
1
|. . .
|Eσ
id
k
σ
m
k
|Qσ
q
|A
1
σ
dist
|. . . |A
l
), where ˜n is the set of
all restricted names, which includes the set of the pro-
tocol’s private channels; Cσ
id
i
σ
a
i
s are the processes
run by the candidates, the substitutions σ
id
i
and σ
a
i
specify the identity and the answers of the i
th
candi-
date respectively; Eσ
id
i
σ
m
i
s are the processes run by
the examiners, the substitution σ
id
i
specifies the i
th
ex-
aminer’s identity, and σ
m
i
specifies for each possible
question/answer pair the corresponding mark; Q is
the process run by the question committee, the substi-
tution σ
q
specifies the exam questions; the A
i
s are the
processes run by the exam authorities, the substitu-
tion σ
dist
determines which answers will be submitted
to which examiners for grading. Without loss of gen-
erality, we assume that A
1
is in charge of distributing
the copies to the examiners.
Definition 2 does not specify whether the examin-
ers are machines or humans. For the purpose of our
model this distinction is not necessary; it is sufficient
that an examiner attributes a mark to a given answer.
Note that Q and A
1
could coincide if for instance there
is only one authority A, in that case we can write sim-
ply Aσ
q
σ
dist
instead of Qσ
q
|A
1
σ
dist
. We organize an
e-exam’s steps in four phases. (1) Registration: the
exam authority (the registrar) creates a new examina-
tion and checks the eligibility of candidates who at-
tempts to register for it. (2) Examination: the exam
authority authenticates the candidates, and sends to
each of them an exam form that contains the exam
questions. Each candidate fills the form with his an-
swer, and submits it to the exam collector. (3) Mark-
ing: the authority distributes the forms submitted by
the candidates to the examiners, who in their turn
evaluate and mark them; (4) Notification: once the
forms have been evaluated, the marks are notified to
the candidates.
3 SECURITY PROPERTIES
We propose a formalization for authentication and
privacy properties. They best represent exam security
requirements as corroborated by other works (Furnell
et al., 1998; Giustolisi et al., 2013; Weippl, 2005).
FormalAnalysisofElectronicExams
103
We introduce four authentication properties meant
to ensure the associations between the candidate’s
identity, the answer, and the mark being preserved
through all phases. When authentication holds there
is no loss, no injection, and in general no manipula-
tion of the exam forms from examination to notifica-
tion. We also introduce five privacy propertiesthat en-
sure the anonymity of critical parties in order to pre-
vent bribing, favouritisms, and to guarantee fairness
among candidates. In the context of e-exams, there
are other classes of properties that might be of inter-
est, such as verifiability, reliability, or accountability,
but we do not study them here. We leave this task as
future work.
3.1 Authentication Properties
We model our authentication properties as correspon-
dence properties, a well-knownapproach (Ryan et al.,
2000; Ryan and Smyth, 2011). Specific events, whose
parameters refer to the pieces of information in the
exam form, flag important steps in the execution of
the exam. Events are annotations that do not change a
process behavior, but are inserted at precise locations
to allow reasoning about the exam’s execution. In the
following id
c is the candidate identity, ques the ques-
tion(s), ans the answer(s), mark the mark(s), id
form
is an identifier of the exam form used during marking,
and id
e is the examiner’s identity.
reg(id
c): is the event inserted into the registrar
process at the location where candidate id
c has
successfully registered for the exam.
submitted(id
c, ques, ans): is the event inserted
into the process of candidate id c in the exami-
nation phase, at the location where he sends his
answer ans corresponding to the question ques.
collected(id c, ques, ans): is the event inserted
into the exam collector’s process in the examina-
tion phase, just after it received and accepted the
exam form (id
c, ques, ans) from candidate id c.
distrib(id
c, ques, ans, id form, id e): is the event
inserted into the authority process in the mark-
ing phase, when it assigns the exam form
(id
c, ques, ans) from candidate id c to the exam-
iner id e using the identifier id form.
marked(ques, ans, mark, id
form, id e): is the
event inserted into the examiner id
es process
in the marking phase, at the location where he
marked the question/answer pair (ques, ans) iden-
tified by id
form with the mark mark.
notified(id
c, mark): is the event inserted into the
process of candidate id c in the notification phase,
just after he received and accepted the mark mark
from the responsible authority.
Note that the id
form is only used to identify an exam
form during marking. This could be a pseudonym to
allow anonymous marking, or simply the candidate
identity if the marking is not anonymous. In exam
with only one examiner the event distrib might ap-
pear to be unnecessary. Yet, it is needed. It links
exam forms to id
form and, for instance, helps reveal-
ing when identical answers are graded multiple times
(and not necessarily with the same mark).
Our events allow us to express authentication
properties as correspondence properties, all having
the following structure: “on every trace the event e
1
is preceded by the event e
2
”. The first authentication
property is Answer Origin Authentication and con-
cerns both the registration and examination phases. It
ensures that only one exam form from each candidate
and only the forms submitted by eligible candidates
(registered) are actually collected.
Definition 3 (Answer Origin Authentication). An
e-exam protocol ensures Answer Origin Authentica-
tion if, for every e-exam process EP, each occurrence
of the event collected(id
c, ques, ans) is preceded by
a distinct occurrence of the event reg(id
c) on every
execution trace.
At examination phase, each candidate submits his
exam form with an answer, and the collector collects
the forms. Form Authorship ensures that the contents
of each collected exam form (id
c, ques, and ans) are
not modified after submission.
Definition 4 (Form Authorship). An e-exam pro-
tocol ensures Form Authorship if, for every e-
exam process EP, each occurrence of the event
collected(id
c, ques, ans) is preceded by a distinct oc-
currence of the event submitted(id c, ques, ans) on ev-
ery execution trace.
Similarly, Form Authenticity ensures that the con-
tent of each exam form is not modified after the col-
lection and until after the form is marked by an exam-
iner.
Definition 5 (Form Authenticity). An e-exam
protocol ensures Form Authenticity if, for ev-
ery e-exam process EP, each occurrence of
the event marked(ques, ans, mark, id
form, id e)
is preceded by a distinct occurrence of the
events distrib(id
c, ques, ans, id form, id e) and
collected(id
c, ques, ans) on every execution trace.
At notification phase, the candidate should receive
the mark which was assigned by the examiner to his
answer. We call this property Mark Authenticity.
SECRYPT2014-InternationalConferenceonSecurityandCryptography
104
Definition 6 (Mark Authenticity). An e-
exam protocol ensures Mark Authenticity if,
for every e-exam process EP, each occur-
rence of the event notified(id
c, mark) is pre-
ceded by a distinct occurrence of the events
marked(ques, ans,mark, id
form, id e) and
distrib(id
c, ques, ans, id form, id e) on every ex-
ecution trace.
Note that Mark Authenticity ensures that the can-
didate is notified with the mark delivered by the ex-
aminer on the answer assigned to him by the author-
ity. This answer may be different from that submitted
by the candidate. Only if also Form Authorship and
Form Authenticity hold then the candidate can be sure
that the assigned and submitted answers are identical.
Mark Authenticity does not guarantee that the mark is
computed correctly.
3.2 Privacy Properties
We model our privacy properties as observational
equivalence, a standard choice for such kind of prop-
erties (Ryan and Schneider, 2001; Ryan and Smyth,
2011). We use the labeled bisimilarity (
l
) to express
the equivalence between two processes (Abadi and
Fournet, 2001). Informally, two processes are equiv-
alent if an observer has no way to tell them apart.
As a notation, we use what in applied π-calculus is
called “context”. The context EP
I
[
] is the process EP
without the identities in the set I; they are replaced by
“holes”. We use it when we need to specify exactly
the processes for candidates id
1
and id
2
without re-
peating the entire e-exam instance. This is done by
rewriting EP as EP
{id
1
,id
2
}
[Cσ
id
1
σ
a
1
|Cσ
id
2
σ
a
2
]. No-
tation EP|
e
denotes the process EP without the code
that follows the event e. The first privacy property
says questions are kept secret until the exam starts.
Definition 7 (Question Indistinguishability). An e-
exam protocol ensures Question Indistinguishability
if for any e-exam process EP that ends with the regis-
tration phase, any questions q
1
and q
2
, we have that:
EP
{id
Q
}
[Qσ
q
1
]|
reg
l
EP
{id
Q
}
[Qσ
q
2
]|
reg
.
Question Indistinguishability states that two pro-
cesses with different questions have to be obser-
vationally equivalent until the end of the registra-
tion phase. This prevents the attacker from ob-
taining information about the exam questions before
the examination phase starts. This property requires
the question committee to be honest; otherwise the
property is trivially violated since the committee re-
veals the questions to the attacker. However, it is
particularly interesting to consider dishonest candi-
dates, as they might be interested in obtaining the
questions in advance. We can do this by replac-
ing honest candidates with dishonest ones. For ex-
ample, if we assume that candidate id
1
is dishon-
est, we obtain EP
{id
1
,id
Q
}
[(Cσ
id
1
σ
a
1
)
c
1
,c
2
|Qσ
q
1
]|
reg
l
EP
{id
1
,id
Q
}
[(Cσ
id
1
σ
a
1
)
c
1
,c
2
|Qσ
q
2
]|
reg
.
The next property ensures that the marking pro-
cess is done anonymously, i.e., that two instances
where candidates swap their answers cannot be dis-
tinguished until after the end of the marking phase.
This may be desirable to ensure fairness of the grad-
ing, and is a requirement in some exam settings (at
some universities or for competitive examinations).
Definition 8 (Anonymous Marking). An e-exam
protocol ensures Anonymous Marking if for any
e-exam process EP that ends with the marking phase,
any two candidates id
1
and id
2
, and any two answers
a
1
and a
2
, we have that:
EP
{id
1
,id
2
}
[Cσ
id
1
σ
a
1
|Cσ
id
2
σ
a
2
]|
mark
l
EP
{id
1
,id
2
}
[Cσ
id
1
σ
a
2
|Cσ
id
2
σ
a
1
]|
mark
.
Anonymous Marking ensures that the process
where id
1
answers a
1
and id
2
answers a
2
is equivalent
to the process where id
1
answers a
2
and id
2
answers
a
1
. This prevents the attacker to obtain the identity
of the candidate who submitted a certain answer be-
fore the marking phase ends. For this property, it is
interesting to consider dishonest examiners. It can be
done using the same technique employed for dishon-
est candidates outlined above. We can also have some
dishonest candidates, however the candidates id
1
and
id
2
who are assigned the two different answers have
to be honest – otherwise the property can be trivially
violated by one of them revealing his answer to the
attacker.
To prevent bribing or coercion of the examiners, it
might be interesting to ensure their anonymity, so that
no candidate knows which examiner marked his copy.
Definition 9 (Anonymous Examiner). An e-exam
protocol ensures Anonymous Examiner if for any
e-exam process EP, any two candidates id
1
, id
2
, any
two examiners id
1
, id
2
, and any two marks m
1
, m
2
, we
have that: EP
{id
1
,id
2
,id
1
,id
2
,id
A
1
}
[Cσ
id
1
σ
a
1
|Cσ
id
2
σ
a
2
|
Eσ
id
1
σ
m
1
|Eσ
id
2
σ
m
2
|A
1
σ
dist
1
]
l
EP
{id
1
,id
2
,id
1
,id
2
,id
A
1
}
[Cσ
id
1
σ
a
1
|Cσ
id
2
σ
a
2
|
Eσ
id
1
σ
m
2
|Eσ
id
2
σ
m
1
|A
1
σ
dist
2
] where σ
dist
1
attributes
the exam form of candidate id
1
to examiner id
1
and
the exam form of candidate id
2
to examiner id
2
, and
σ
dist
2
attributes the exam form of candidate id
1
to
examiner id
2
and the exam form of candidate id
2
to
examiner id
1
.
Anonymous Examiner ensures that a process in
which examiner id
1
grades the exam form of candi-
date id
1
and examiner id
2
grades that of candidate id
2
FormalAnalysisofElectronicExams
105
cannot be distinguished from a process in which id
1
grades the exam form of id
2
and id
2
grades that of id
1
.
Note that to ensure that in both cases the candidates
receive the same mark, we also have to swap σ
m
1
and
σ
m
2
between the examiners. Similar to Anonymous
Marking, this property prevents the attacker to obtain
or guess the identity of the examiner who marked a
certain answer. Anonymous Examiner requires that
the examiners id
1
and id
2
are honest, otherwise it will
trivially violated by one of them revealing the mark
he gave. We can again include dishonest candidates
as they might be interested in finding out which ex-
aminer marked their copies.
In some exams settings the marks have to remain
private. This is formalized in the next property.
Definition 10 (Mark Privacy). An e-exam protocol
ensures Mark Privacy if for any e-exam process EP,
any marks m
1
, m
2
, we have that: EP
{id
}
[Eσ
id
σ
m
1
]
l
EP
{id
}
[Eσ
id
σ
m
2
].
Mark Privacy guarantees that two processes
where the examiner id
1
assigns for the same answer,
entailed by the same context EP, two different marks
m
1
, m
2
, cannot be distinguished from each other. De-
pending on the exam policy this can be an optional
property since some exams system may publicly dis-
close the marks of the candidates. However, the intu-
ition here is that candidate’s performance should not
be known to any other candidate. Again, we can as-
sume that some candidates are dishonest and try to
find out the marks of their colleagues, or that an ex-
aminer tries to find out the mark achieved by a can-
didate. The candidate who is assigned the two differ-
ent marks has to be honest otherwise the property
is violated by him revealing his mark to the attacker.
Similarly the examiner assigning the marks has to be
honest, otherwise he can reveal the mark himself.
The previous definition of Mark Privacy ensures
that the attacker cannot know the mark of a candidate.
A weaker variant of Mark Privacy is Mark Anonymity,
i.e., the attacker might know the list of all marks, but
is unable to associate a mark to its corresponding can-
didate. This is often the case in practice, where a list
of pseudonyms (e.g., student numbers) and marks is
published.
Definition 11 (Mark Anonymity). An e-exam pro-
tocol ensures Mark Anonymity if for any e-exam pro-
cess EP, any candidates id
1
, id
2
, any examiner id
1
,
any answers a
1
, a
2
and a distribution σ
dist
that as-
signs the answers of both candidates to the exam-
iner, and two substitutions σ
m
a
and σ
m
b
which are
identical, except that σ
m
a
attributes the mark m
1
to
the answer a
1
and m
2
to a
2
, whereas σ
m
b
attributes
m
2
to the answer a
1
and m
1
to a
2
, we have that:
EP
{id
1
,id
2
,id
1
,id
A
1
}
[Cσ
id
1
σ
a
1
|Cσ
id
2
σ
a
2
|Eσ
id
1
σ
m
a
|A
1
σ
dist
]
l
EP
{id
1
,id
2
,id
1
,id
A
1
}
[Cσ
id
1
σ
a
1
|Cσ
id
2
σ
a
2
|Eσ
id
1
σ
m
b
|A
1
σ
dist
]
The definition states that if an examiner id
1
, who
is assigned the same answers a
1
and a
2
as σ
dist
is un-
changed, swaps the marks between these answers, the
two situations cannot be distinguished by the attacker.
This means that a list of marks can be public, but the
attacker must be unable to link the marks to the can-
didates. Again, we can consider dishonest parties, but
this definition requires the two concerned candidates
and the two concerned examiners to be honest. Other-
wise they can simply reveal the answer and the asso-
ciated mark, which allows to distinguish both cases.
It is also easy to see that a protocol ensuring Mark
Privacy also ensures Mark Anonymity. In fact, σ
m
a
and σ
m
b
are special cases of σ
m
1
and σ
m
2
.
4 Huszti & Peth
˝
o PROTOCOL
We first analyze the Huszti & Peth˝o protocol (Huszti
and Peth˝o, 2010), which we call concisely H&P pro-
tocol. It aims to ensure authentication and privacy for
e-exams in presence of dishonest candidates, exam-
iners and exam authorities. The original paper also
presents an informal security analysis based on con-
jectures. Such conjectures contribute to our motiva-
tion on the need of a framework for the formal anal-
ysis of e-exams. Notably, all the messages within the
H&P protocol are sent via a reusable anonymous re-
turn channel (RARC) (Golle and Jakobsson, 2003) to
achieve privacy properties.
A RARC implements anonymous
4
two-way con-
versations. It allows the party that initiates the pro-
tocol to send an anonymous message to a recipient.
The recipient can reply without learning the sender’s
identity, but knowing that his reply will be dispatched
to the actual sender. The entire conversation re-
mains untraceable to an external attacker. A RARC
is implemented by a re-encryption mixnet. The mix
servers jointly generate and share an ElGamal key
pair (PK
MIX
, SK
MIX
) and a pair of public/private sign-
ing keys (SPK
MIX
, SSK
MIX
). The sender A and the re-
ceiver B also hold ElGamal public/private key pairs,
(PK
A
, SK
A
) and (PK
B
, SK
B
) respectively. A and B are
represented by ID
A
and ID
B
, identity tags which can
be for example As and Bs email addresses.
To send the message m to B, the agent A sub-
mits to the mixnet the tuple Mix(m, A, B) that de-
notes ({ID
A
, PK
A
}
PK
MIX
, {m}
PK
MIX
, {ID
B
, PK
B
}
PK
MIX
) and
4
Note that although the original security definition re-
quires anonymity of the messages, it does not require se-
crecy of the messages.
SECRYPT2014-InternationalConferenceonSecurityandCryptography
106
proves knowledge of {ID
A
, PK
A
} and of {ID
B
, PK
B
}.
The proofs are meant to avoid that the attacker de-
crypts the triplet content by using the mixnet as
a decryption oracle (in Section 4.2 we prove this
claim to be false considering a Dolev-Yao threat
model). The mixnet waits to collect more triplets
and then shuffles them. Then, it adds a checksum
to the triplets in order to guarantee their integrity
while they are shuffled, yet this provides no end-to-
end integrity protection. The message m is then re-
encrypted with the public key of B using a switch-
ing encryption keys technique. The mixnet signs
the encrypted public key of A. Thus B receives the
pair (sign({ID
A
, PK
A
}
PK
MIX
, SK
MIX
), {m}
PK
B
) where
sign(x, sk) is message x plus the signature with
the secret key sk. Then B replies to A with
a new message m
by sending to the mixnet
(Mix(m
, B, A),sign({ID
A
, PK
A
}
PK
MIX
, SK
MIX
)) and prov-
ing only knowledge of {ID
B
, PK
B
}. The mixnet
checks the proof and the signature, and then processes
the tuples like a normal message.
4.1 Protocol Description
We use ProVerif to verify the protocol. The ProVerif
model is based on the description presented in orig-
inal paper (Huszti and Peth˝o, 2010). Here we only
give an overview of the protocol.
The H&P protocol relies upon different crypto-
graphic building blocks. The ElGamal cryptosys-
tem (Elgamal, 1985) is used to provide parties
with public/private key pairs. A RARC implements
anonymous two-way communication. A network
of servers provides a timed-release service (NET),
which contributes to create and revoke the candi-
date’s pseudonym. More precisely, their contribution
to the pseudonym is shared among the servers using
the threshold Shamir secret sharing system (Shamir,
1979). At notification, a subset of the NET servers
can use their shares to recover the secret and de-
anonymize the pseudonymof the candidate. Then, the
exam authority can associate the answer with the cor-
responding candidate. To avoid plagiarism, the proto-
col assumes that no candidate revealsits private key to
another candidate and that invigilators supervise can-
didates during the examination. We describe the pro-
tocol in ve phases, distinguishing the examiner and
the candidate registration. However, according to our
model, the two registration (sub-)phases are merged
into a single phase.
Examiner Registration: The exam authority pub-
lishes the public parameters to identify a new exami-
nation. The question committee then signs and sends
the questions and the starting time of the phases en-
crypted with the public key of the RARC mixnet. The
mixnet forwards the message only when the exami-
nation begins. The examiner is then provided with a
pseudonym, which is jointly generated by the exam
authority and the examiner. The examiner verifies
the correctness of the pseudonym by using a zero-
knowledge proof (ZKP). Then, the examiner sends
his pseudonym to the exam authority, and proves the
knowledge of his secret key.
Candidate Registration: The registration of a can-
didate slightly differs from the registration of an ex-
aminer. The candidate pseudonym is jointly calcu-
lated by the exam authority, the candidate, and also
the NET to provide anonymityfor the candidates. The
NET stores the secret values used for the pseudonym
generation, which can be used to de-anonymize the
candidate after the examination has finished. Again,
the candidate finally verifies the correctness of his
pseudonym using a ZKP.
Examination: The candidate sends his pseudonym
via the RARC to the exam authority and proves the
knowledge of his private key. Then, the exam author-
ity checks whether the candidate is registered for the
examination, and sends him the questions signed by
the question committee. The candidate sends his an-
swer, again via the RARC. The exam authority replies
with a receipt which consists of the hash of all param-
eters seen by the exam authority during the examina-
tion, the transcription of the ZKPs, and the time when
the answer was submitted.
Marking: The exam authority chooses an exam-
iner who is eligible for the examination, and forwards
him the answer via the RARC. Then the examiner as-
signs a mark to the answer, and authenticates them
using a ZKP.
Notification: When all the answers are marked,
the NET de-anonymizesthe pseudonymslinked to the
answers, and the exam authority stores the marks.
4.2 Formal Analysis
The equational theory depicted in Table 1 models the
cryptographic primitives used within the H&P pro-
tocol. The equational theory includes well-known
models for probabilistic encryption and digital signa-
tures. Inspired by Backes et al. (Backes et al., 2008b),
we model the ZKP of knowledge of a secret expo-
nent as two functions, proof and verification. The
proof function zkp
proof(public, secret) takes as ar-
guments a secret and public parameters (i.e. the ex-
ponent and the generator to the power of the expo-
nent). It can be constructed only by the prover who
knows the secret parameter. The verification function
zkpsec(zkp
proof(public, secret), verinfo) takes as ar-
FormalAnalysisofElectronicExams
107
guments the proof function and the verification pa-
rameter verinfo. The verifier only accepts the proof
if the relation between verinfo and secret is satis-
fied. However, we support the model for the ZKP
of the equality of discrete logarithms zkp
proof with
tables in ProVerif. This is due to the difficulties of
ProVerif when dealing with associativity of multiple
exponents, which is used in the H&P protocol. We
also assume the same generator is used for generat-
ing the pseudonyms of candidates and examiners, in
order to avoid non-termination in ProVerif. This is
sound because we distinguish the roles, and each prin-
cipal is identified by its public key. We replace the
candidate identity with his corresponding pseudonym
inside the events to check authentication properties.
We note that the replacement is also sound because
the equational theory preserves the bijective mapping
between the keys that identify the candidate and his
pseudonym.
Table 1: Equational theory to model H&P protocol.
decrypt(encrypt(m, pk(k), r), k) = m
getmess(sign(m, k)) = m
checksign(sign(m, k), pk(k)) = m
exp(exp(g, x),y) = exp(exp(g, y), x)
checkproof(xproof (p, p1, g, exp(g, e),e),
p, p1, g, exp(g, e)) = true
zkpsec(zkp
proof (exp(b, e), e),exp(b, e)) = true
First we analyzed the RARC alone and found an
attack on anonymity and privacy, which is detailed in
the next paragraph. We then replaced the RARC with
an implementation of a secure RARC using honest
parties to check if the protocol ensures some proper-
ties given a working anonymouschannel. In this case,
ProVerif terminates for all properties on H&P
5
.
The result of the verification together with the
time required for ProVerif to conclude on a standard
PC (Intel i7, 8GB RAM), are summed up in Table 2.
Attack on RARC: ProVerif shows that the RARC
fails to guarantee both secrecy of messages and
anonymity of sender and receiver identities, which is
its main purpose inside the H&P protocol. We refer
the triplet hc
1
, c
2
, c
3
i as the encrypted messages that A
submits to the mixnet when she wants to send a mes-
sage to B. From the description of RARC given at the
beginning of this section, we recall that c
1
encrypts
the As public key, c
2
encrypts the message to B, and
5
All ProVerif codes are available on line
http://apsia.uni.lu/stast/codes/exams/
proverif
code secrypt.tar.gz
Table 2: Summary of our analysis on the formal model of
the H&P protocol.
Property Result Time
Answer Origin Authentication × < 1 s
Form Authorship × < 1 s
Form Authenticity × < 1 s
Mark Authenticity × < 1 s
Question Indistinguishability × < 1 s
Anonymous Marking × 8 m 46 s
Anonymous Examiner × 9 m 8 s
Mark Privacy × 39 m 8 s
Mark Anonymity × 1h 15 m 58 s
c
3
encrypts the Bs public key. All cipher-texts are
encrypted with the mixnet’s public key.
The attacker in control of the network can use
the RARC as a decryption oracle, letting the RARC
reveal any of the plaintexts. The attack works as
follows. The attacker chooses one of the three ci-
phertexts (depending on whether he wants to tar-
get the contents of the message, or the identities of
the sender and receiver) and submits this as a new
message. For example, if the attacker targets c
1
=
{ID
A
, PK
A
}
PK
MIX
, he resubmits c
1
as a new encrypted
message, which means that c
2
= c
1
in the new triplet.
He can leave the encryption of the senders key and
the proof concerning the key unchanged, but replaces
the encryption of the receiver’s key with a public key
PK
I
for which he knows the corresponding secret key
SK
I
. In our example this means c
3
= {ID
I
, PK
I
}
PK
MIX
.
The attacker can also provide the necessary proof of
knowledge of plaintext, since he knows this plaintext.
The RARC then mixes the input messages, and
sends the encryption of the message under the re-
ceiver’s public key to the receiver. In our example the
attacker receives {ID
A
, PK
A
}
PK
I
. Since the attacker
knows the secret key SK
I
he can obtain the original
message. In our example he gets ID
A
, the identity of
the sender which should have remained anonymous.
Since the attacker can substitute any of the items in
the triplet as the new message, the RARC does neither
ensure secrecy of the messages nor the anonymity of
the sender or the receiver. Note that the checksum
meant to guarantee the integrity of the triplet is only
added after the submission of the message and is only
used inside the mixnet. Hence, the checksum does
not prevent the attacker from submitting a modified
triplet. Even if it were added before, it would not pre-
vent the attack as the knowledge of the ciphertexts is
sufficient to compute the checksum.
Note that the RARC was originally designed to
withstand a passive attacker, which however is al-
lowed to statically corrupt parties. We argue that this
is not realistic in the e-exam setting where dishonest
parties could try to cheat. Moreover, even static cor-
SECRYPT2014-InternationalConferenceonSecurityandCryptography
108
ruption is sufficient to carry out the attack outlined
above: one dishonest party that can send and receive
messages via the RARC is sufficient. Also, he has
to intercept the message before it enters the RARC,
but this is difficult to prevent in a normal unsecured
network such as the Internet.
Authentication properties: We verified the au-
thentication properties modelling the RARC as an
ideal anonymous (yet not secret, according to the
original security definition) channel. Note that the
following attacks also remain valid if the protocol
adopts the RARC.
ProVerif finds an attack on Answer Origin Au-
thentication where the attacker can create a fake
pseudonym that allows him to take part in an exam
for which he did not registered. This is possible be-
cause the exam authority does not check whether the
pseudonym has been actually created using the partial
information provided by the time-release service. The
attacker generates his own secret key SK
A
, and cal-
culates an associate pseudonym, which sends to the
exam authority. The exam authority successfully veri-
fies the received data and that the attacker knows SK
A
,
thus the exam authority accepts the answer. Form Au-
thorship fails due to the same attack: in fact, the exam
authority may collects an exam form which is modi-
fied by changing the pseudonym to a one chosen by
the attacker.
ProVerif also shows that the H&P protocol does
not ensure Form Authenticity, because there is no
mechanism that allows the examiner to check whether
the answers have been forwarded by the exam author-
ity. Even if the original RARC is used and the answer
is encrypted with the public key of the mixnet, this
does not guarantee that the exam authority actually
sent the message.
ProVerif provides a similar attack for Mark Au-
thenticity. In fact, the attacker can forward any an-
swer to any examiner, even if the answer was not col-
lected by the exam authority. Morevoer, the attacker
can notify the candidate by himself with a mark of his
choice.
Privacy properties: ProVerif finds an attack on
Question Indistinguishability. This is because the at-
tack on the RARC exposes the message and the iden-
tities of the sender and receiver. As the questions
are sent through the RARC, the attacker can obtain
them. Moreover, as the candidate’s answer is also
sent through the RARC, the protocol does not ensure
Anonymous Marking: the answer can be linked to its
corresponding sender. The protocol ensures neither
Mark Privacy nor Anonymous Examiner, as the marks
are also sent through the RARC. Hence, they can be
decrypted and the examiner can be identified.
We checked the H&P protocol in ProVerif assum-
ing correct RARC (i.e. ensuring anonymity, but no
secrecy). Also in this case ProVerif shows an at-
tack for each property. Anonymous Examiner can
be violated because the attacker can track which ex-
aminer accepts the ZKP when receiving the partial
pseudonym, and then associate to the examiner the
answer that the latter grades. Moreover, a similar at-
tack on Anonymous Marking remains: the attacker
can check whether a candidate accepts the ZKP to
associate him with a pseudonym, and then identify
his answer. Finally, neither Mark Privacy nor Mark
Anonymity are ensured because the examiner sends
the mark to the exam authority in clear.
To sum up, the H&P protocol ensures no proper-
ties at all, and ProVerif discovers attacks. While the
authentication properties fail due to a weak protocol
design, the privacy properties fail because of an in-
appropriate use of the RARC, which was neither de-
signed to ensure secrecy nor to withstand active at-
tackers. Moreover, we identified flaws in the RARC
even in a static corruption setting, and in the H&P
protocol assuming a correct anonymous channel.
5 Remark! PROTOCOL
We first give the protocol presented in (Giustolisi
et al., 2014) and then the results of our analysis.
5.1 Protocol Description
The Remark! protocol has the same set of parties of
the H&P protocol, but relies on a different approach.
The NET is indeed several servers that implement an
exponentiation mixnet (Haenni and Spycher, 2011).
The speciality of exponentiation mixnets is that each
server blinds its entries by a common exponent value.
On entry X, the mixnet outputs X
r
where r is the prod-
uct of the secret exponent values of the servers. At
registration, the NET creates the pseudonyms for the
candidates and examiners without involving any of
them. The pseudonyms are eventually used as public-
key encryption and signature verification keys in such
a way to allow parties to communicate anonymously.
A bulletin board
6
is used to publish the pseudonyms,
the test questions and the receipts of test submissions.
The combination of the exponentiation mixnet and
a bulletin board allows the protocol to not rely on a
RARC as anonymous channel.
6
A public append-only memory.
FormalAnalysisofElectronicExams
109
Remark! only assumes that each party is given a
pair of public/private key with a common generator
g, i.e. the private key x and the public key y = g
x
.
Below, we present the protocol within the four e-exam
phases.
Registration: The list of eligible candidates’ and
examiners’ public keys is sent as a batch to the NET.
The NET calculates the pseudonyms by raising the
initial public keys to a common value r =
i
r
i
. More
specifically, each mix server raises the input message
to a secret value r
i
, and forwards it to another mix
server. At the same time the NET blindly permutes
the batch of public keys. The so obtained keys even-
tually become the pseudonyms for candidates and ex-
aminers. Along with the pseudonyms y
= y
r
= (g
x
)
r
,
the NET publishes a new generator h, which is the
output of g raised to the product of each mix server
secret value, i.e. h = g
r
. Both the candidates and
the examiners can identify their own pseudonyms by
raising h to their secret key x, i.e. h
x
= (g
r
)
x
. The
pseudonyms from now on serve as public encryption
and signature verification keys. Two different batches
are used for candidates and examiners because only
the identities of candidates are revealed at notifica-
tion.
Examination: The exam authority signs and
encrypts the test questions with the candidate’s
pseudonym and publishes them on the bulletin board.
Each candidate submits his answer, which is signed
with the candidate’s private key (but using the gen-
erator h instead of g) and encrypted with the public
key of the exam authority. The exam authority col-
lects the test answer, checks its signature using the
candidate’s pseudonym, re-signs it, and finally pub-
lishes its encryption with the corresponding candi-
date’s pseudonym as receipt.
Marking: The exam authority encrypts the signed
test answer with an eligible examiner pseudonym and
publishes the encryption on the bulletin board. The
corresponding examiner marks the test answer, and
signs it with his private key (again using the generator
h instead of g). The examiner then encrypts it with
the exam authority public key, and submits its marks
to the exam authority.
Notification: When the exam authority receives
all the candidate evaluations, it publishes the signed
marks, each encrypted with the corresponding can-
didate’s pseudonym. Then, the NET servers de-
anonymize the candidate’s pseudonyms by reveal-
ing their secret exponents. Hence the candidate
anonymity is revoked, and the mark can finally be
registered. Note that the examiner’s secret exponent
is not revealed to ensure his anonymity even after the
exam concludes.
checkpseudo(pseudo
pub(pk(k), rce),
pseudo
priv(k, exp(rce))) = true
decrypt(encrypt(m, pk(k), r), k) = m
decrypt(encrypt(m, pseudo
pub(pk(k),
rce), r), pseudo
priv(k, exp(rce))) = m
getmess(sign(m, k)) = m
checksign(sign(m, k), pk(k)) = m
checksign(sign(m, pseudo
priv(k,
exp(rce))), pseudo
pub(pk(k), rce)) = m
Table 3: Equational theory to model Remark! protocol
5.2 Formal Analysis
We analyze Remark! with ProVerif, following similar
techniques as the one used in the analysis of the H&P
protocol. Table 4 sums up the results together with the
time required for ProVerif to conclude on the same PC
used for H&P. We model the bulletin board as a pub-
lic channel, and use the equational theory depicted in
Table 3. The equations for encryption and signatures
are standard, but we also added the possibility of us-
ing the pseudonym keys to encryptor sign. The public
pseudonym, which also serves as exam form identi-
fier, is obtained using the function pseudo
pub on the
public key and the random exponent. The function
pseudo
priv can be used to decrypt or sign messages,
using the private key and the new generator g
r
(mod-
elled using the function exp) as parameters. The func-
tion checkpseudo allows us to check if a pseudonym
corresponds to a given secret key (or its pseudonym
variant).
Authentication properties: Supposing an attacker
in control of the network and all parties to be honest,
we can successfully verify all authentication proper-
ties in ProVerif. For this, we replace the candidate
identity with the candidate’s pseudonym inside the
events. This is sound as each candidate is uniquely
identified by his keys, and there is a bijective mapping
between keys and pseudonyms by construction of the
equational theory (for a given random exponent).
We also verified the authentication properties con-
sidering dishonest parties. In this case, all properties
are guaranteed except Form Authenticity. The attack
trace shows that a dishonest candidate can pick the
examiner of his choice by re-encrypting the signed
receipt received from the exam authority. It means
that the candidate can influence the choice of the ex-
aminer who will correct his exam. As the protocol
description envisages an access control for publish-
ing into the bulletin board, a feature that we could not
SECRYPT2014-InternationalConferenceonSecurityandCryptography
110
Table 4: Summary of our analysis on the formal model of
the Remark! protocol. (
) Form Authenticity fails with dis-
honest candidate. It holds after applying our fix.
Property Result Time
Answer Origin Authentication X < 1 s
Form Authorship X < 1 s
Form Authenticity X
< 1 s
Mark Authenticity X < 1 s
Question Indistinguishability X < 1 s
Anonymous Marking X 2 s
Anonymous Examiner X 1 s
Mark Privacy X 3 m 32 s
code in ProVerif, we cannot claim this to be an attack
as the candidate may not be allowed to post on the
bulletin board. However, we demonstrate that with a
simple fix there is no need of access control policies
for publishing into the bulletin board. The fix consists
in making the intended pseudonymof an examiner ex-
plicit within the signature that designates the exam-
iner as evaluator of an exam. In doing so, the exam
authoritys signature within the receipt cannot be used
by a candidate to designate any examiner because the
receipt includes no examiner’s pseudonym. The exam
authority will only accept exam evaluations that con-
tain its signature on examiner’s pseudonym. Consid-
ering the fix, ProVerif confirms that Remark! guaran-
tees all the security propertiesincluding Form Authen-
ticity, even in presence of dishonest parties.
Privacy properties: All the privacy properties are
satisfied. For Question Indistinguishability, we only
assume the exam authority to be honest, and then con-
clude that the property holds. For Mark Privacy, we
assume only the concerned candidate and examiner,
as well as the exam authority, to be honest. All other
candidates and examiners are dishonest, and ProVerif
still concludes successfully. Note that this subsumes
a case with multiple honest candidates and examin-
ers, since a dishonest party can behave like an hon-
est party. This also implies that the protocol ensures
Mark Anonymity as noted above. For Anonymous Ex-
aminer, we assume only the examiners and the NET
to be honest. If the NET publishes the pseudonyms in
random order, ProVerif concludes successfully. Sim-
ilarly for Anonymous Marking, we assume only the
candidates and the NET to be honest. Again, if
the NET publishes the pseudonyms in random order,
ProVerif concludes successfully.
6 CONCLUSION
We define the first formal framework for the analysis
of secure e-exam protocols. We show how to model
e-exam protocols in the applied π-calculus, and define
nine relevant security properties: four authentication
properties and five privacy properties.
Using ProVerif, we analyze the security of two
e-exam protocols. The first protocol has only been
argued to be secure. Our analysis shows that it in-
deed satisfies none of the nine properties. Authenti-
cation is compromised because of inaccuracies in the
protocol design, whereas most of attacks invalidating
privacy exploit attacks on the RARC. These attacks
compromise secrecy and anonymity of the messages,
and exploit the absence of a proof of knowledge of
the submitted message to the RARC, which allows
its use as a decryption oracle. Such a proof is not
explicitly required in the original specification of the
RARC, and is certainly missing in the H&P protocol:
the “exam authority” is required to forward questions
and answers without knowing them, and thus cannot
prove knowledge of them when submitting them to
the RARC. Even when assuming a perfect RARC en-
suring anonymity, we still have attacks on all proper-
ties. Thus, we think that fixing the RARC is not suf-
ficient – the protocol requires fundamental changes.
Also Remark!, the second protocol analyzed, has
been only informally argued to be secure in the orig-
inal paper. It presents a weakness concerning Form
Authenticity. We propose a fix and formally verify
that the (fixed) protocol satisfies all the properties
herein considered.
Generally speaking, our framework and our anal-
ysis bring e-exams into the attention of the security
community. E-exams and in general computer-based
assessment tools are becoming widespread, some of
them supported by e-learning platforms such as the
massive open online courses (MOOC). Nevertheless,
they call for being formally proved secure, since most
of them have not been submitted to any rigorous se-
curity analysis. Such applications are complex and
exposed to unprecedented cheating attacks very sub-
tle to be discovered. We set the first research step on
the formal understanding of such systems and estab-
lishes a framework for the automatic analysis of their
security properties.
As a future work we intend to analyze more pro-
tocols designed for computer-based tests although ob-
taining protocol’s specifications from the providers is
not an easy task. Other interesting research works in-
clude the study of the relation between our security
properties as well as the definition of novel properties
such as verifiability, reliability, and accountability.
FormalAnalysisofElectronicExams
111
REFERENCES
Abadi, M. and Fournet, C. (2001). Mobile values, new
names, and secure communication. In POPL. ACM.
Arapinis, M., Bursuc, S., and Ryan, M. (2013). Privacy-
supporting cloud computing by in-browser key trans-
lation. Journal of Computer Security, 21(6):847–880.
Backes, M., Hritcu, C., and Maffei, M. (2008a). Automated
verification of remote electronic voting protocols in
the applied pi-calculus. In CSF. IEEE.
Backes, M., Maffei, M., and Unruh, D. (2008b). Zero-
knowledge in the applied pi-calculus and automated
verification of the direct anonymous attestation proto-
col. In IEEE S & P’08.
Bella, G., Costantino, G., Coles-Kemp, L., and Ric-
cobene, S. (2011). Remote management of face-to-
face written authenticated though anonymous exams.
In CSEDU (2), pages 431–437. SciTePress.
Blanchet, B. (2001). An efficient cryptographic protocol
verifier based on prolog rules. In CSFW. IEEE.
Blanchet, B., Abadi, M., and Fournet, C. (2008). Auto-
mated verification of selected equivalences for secu-
rity protocols. J. Log. Algebr. Program., 75(1):3–51.
Castell`a-Roca, J., Herrera-Joancomart´ı, J., and Dorca-Josa,
A. (2006). A secure e-exam management system. In
ARES, pages 864–871. IEEE Computer Society.
Copeland, L. (2013). School cheat-
ing scandal shakes up atlanta.
http://www.usatoday.com/story/news/nation/2013/04/
13/atlanta-school-cheatring-race/2079327/.
Delaune, S., Kremer, S., and Ryan, M. (2006a). Verifying
properties of electronic voting protocols. In Proceed-
ings of WOTE’06.
Delaune, S., Kremer, S., and Ryan, M. (2009). Verifying
privacy-type properties of electronic voting protocols.
Journal of Computer Security, 17(4):435–487.
Delaune, S., Kremer, S., and Ryan, M. D. (2006b).
Coercion-resistance and receipt-freeness in electronic
voting. In CSFW’06. IEEE.
Dolev, D. and Yao, A. C. (1983). On the security of pub-
lic key protocols. Information Theory, IEEE Transac-
tions on, 29(2):198–208.
Dong, N., Jonker, H. L., and Pang, J. (2010). Analysis of a
receipt-free auction protocol in the applied pi calculus.
In FAST’10, volume 6561 of LNCS. Springer.
Dreier, J., Jonker, H., and Lafourcade, P. (2013a). Defin-
ing verifiability in e-auction protocols. In ASIACCS,
pages 547–552. ACM.
Dreier, J., Lafourcade, P., and Lakhnech, Y. (2011). Vote-
independence: A powerful privacy notion for voting
protocols. In FPS, volume 6888 of LNCS. Springer.
Dreier, J., Lafourcade, P., and Lakhnech, Y. (2012a). Defin-
ing privacy for weighted votes, single and multi-voter
coercion. In ESORICS, LNCS. Springer.
Dreier, J., Lafourcade, P., and Lakhnech, Y. (2012b). A
formal taxonomy of privacy in voting protocols. In
ICC, pages 6710–6715. IEEE.
Dreier, J., Lafourcade, P., and Lakhnech, Y. (2013b). For-
mal verification of e-auction protocols. In POST, vol-
ume 7796 of LNCS, pages 247–266. Springer.
Elgamal, T. (1985). A public key cryptosystem and a signa-
ture scheme based on discrete logarithms. Information
Theory, IEEE Transactions on, 31(4):469–472.
Furnell, S., Onions, P., Knahl, M., Sanders, P., Bleimann,
U., Gojny, U., and R¨oder, H. (1998). A security frame-
work for online distance learning and training. Inter-
net Research, 8(3):236–242.
Giustolisi, R., Lenzini, G., and Bella, G. (2013). What se-
curity for electronic exams? 8th Int. Conf. on Risk and
Security of Internet and Systems (CRiSIS).
Giustolisi, R., Lenzini, G., and Ryan, P. (2014). Remark!:
A secure protocol for remote exams. In Security
Protocols XXII, LNCS. Springer. to appear. Draft
http://apsia.uni.lu/stast/codes/exams/preSPW14.pdf.
Golle, P. and Jakobsson, M. (2003). Reusable anonymous
return channels. In Proc. of the 2003 ACM workshop
on Privacy in the electronic society, WPES ’03. ACM.
Haenni, R. and Spycher, O. (2011). Secure internet voting
on limited devices with anonymized dsa public keys.
In WOTE’11. USENIX.
Herrera-Joancomart´ı, J., Prieto-Bl´azquez, J., and Castell`a-
Roca, J. (2004). A secure electronic examination pro-
tocol using wireless networks. In ITCC. IEEE.
Hjeltnes, T. and Hansson, B. (2005). Cost Effectiveness and
Cost Efficiency in E-learning. QUIS - Quality, Inter-
operability and Standards in e-learning, Norway.
Huszti, A. and Peth˝o, A. (2010). A secure electronic
exam system. Publicationes Mathematicae Debrecen,
77:299–312.
Ryan, M. and Smyth, B. (2011). Applied pi calculus. In
Formal Models and Techniques for Analyzing Security
Protocols, chapter 6. IOS Press.
Ryan, P. Y. A. and Schneider, S. A. (2001). Process algebra
and non-interference. J. Comput. Secur., 9(1-2).
Ryan, P. Y. A., Schneider, S. A., Goldsmith, M., Lowe,
G., and Roscoe:, A. W. (2000). The Modelling and
Analysis of Security Protocols: The CSP Approach.
Addison-Wesley Professional, first edition.
Shamir, A. (1979). How to share a secret. Commun. ACM,
22(11):612–613.
Watson, R. (2014). Student visa system fraud exposed
in BBC investigation. http://www.bbc.com/news/uk-
26024375.
Weippl, E. (2005). Security in E-learning, volume 6 of Ad-
vances in Information Security. Springer Science +
Business Media.
SECRYPT2014-InternationalConferenceonSecurityandCryptography
112